面對未來可能發生的缺水危機，需要對水文地質條件與地下水流動特性有著充分瞭解。水文地質參數如滲透係數及蓄水係數等，乃是地下水數值模擬及水資源管理中必不可少的關鍵資料，但傳統上水文地質參數乃是通過現地抽水實驗的方式獲取，在空間解析度方面存在一定限制。 因此，本研究以時空資料探勘方法—擴展經驗正交函數法(Extended Empirical Orthogonal Function, EEOF)為資料分析基礎結合機器學習模型，以屏東平原為例，建立水文參數推估模式。本研究首先收集了研究區域内地下水水位資料與其他水文資料，透過EEOF分解出時空特徵，並利用其空間特徵結合其他參數建立機器學習模型來推估蓄水係數及滲透係數。結果顯示地下水位進行EEOF分解後，首先可很好提供降雨與地下水之間的時空變異特性資訊；其次，地下水位空間特徵在接下來的水文地質參數推估中，能夠提供非常重要的訊息。最後的模式結果證明，結合了地下水位空間特徵的水文地質參數推估模式所得到結果的準確率較高，期望能為未來臺灣各地區地下水研究提供參考。 ;To face the likelihood and severity of water shortage, understanding the hydrogeological conditions and groundwater flow are important. Hydrogeological parameters of aquifer such as hydraulic conductivity (K) and storage coefficient (S) are the essential and crucial basic data in the groundwater modeling and resource assessment. However, traditionally, the estimation of hydrogeological parameters is inefficient in spatial resolution, time consuming and expensive from pumping test. In this study, a data mining framework based on Extended Empirical Orthogonal Function (EEOF) combing with Machine Learning models were applied to estimate the hydraulic conductivity and storage coefficient. We extract the major spatial-temporal patterns of groundwater level variation from EEOF and use them to build a spatial machine learning model to estimate the hydrogeological parameters in Pingtung plain. The EEOF results have shown that this analysis framework can provide the information of main variation of spatial temporal feature between rainfall and groundwater, and it can provide crucial information for the machine learning model. The model results have shown that the model precision is quite high. This framework could also apply to other aquifers and provide as a very useful information for groundwater modeling and management through the pure data-driven techniques proposed by this study.
